Section 7
Section 7
Section 7
10 1998-06-30 01:54:05-07 drichard Exp ODE Architect Workbook Page 207 on June 30, 1998 at 3:06
t-x
1.5
0.5
x
-0.5
-1
-1.5
0 2 4 6 8 10
t
208 Chapter 11
where r1 is the larger of the two roots r1 and r2 of the indicial equation.
Consult the references for The coefficients an can be determined in the same way as in the earlier
detailed instructions on how to example: differentiate twice, substitute the series for qx1 , px1 and x1 into the
find the coefficients an .
given differential equation, and then find a recurrence formula.
Here are a few things to keep in mind when finding a Frobenius series.
1. The roots of the indicial equation may not be integers, in which case the
series representation of the solution would not be a power series, but is
still a valid series.
2. If r1 r2 is not an integer, then the smaller root r2 of the indicial equa-
tion generates a second solution of the form
x2 (t) = (t t0 )r2 bn (t t0 )n
n=0
which is linearly independent of the first solution x1 (t).
3. When r1 r2 is an integer, a second solution of the form
x2 (t) = Cx1 (t) ln(t t0 ) + bn (t t0 )n+r2
n=0
exists, where the values of the coefficents b n are determined by finding
a recurrence formula, and C is a constant. The solution x 2 (t) is linearly
independent of x1 (t).
Id: chapter11.tex,v 1.10 1998-06-30 01:54:05-07 drichard Exp ODE Architect Workbook Page 209 on June 30, 1998 at 3:06
Bessel Functions
For any nonnegative constant p, the differential equation
t 2 x (t) + tx (t) + (t2 p2 )x(t) = 0
is known as Bessels equation of order p, and its solutions are the Bessel
functions of order p. In normalized form, Bessels equation becomes
If t is very large, Bessels
1
2
t p2
equation looks like the harmonic x (t) + x (t) + x(t) = 0
oscillator equation, x + x = 0. t t2
From this we can see that tp(t) = 1 and t 2 q(t) = t2 p2 , so that tp(t) and
t 2 q(t) are analytic at t0 = 0. Therefore zero is a regular singular point and,
using equation (8), we find that the indicial equation is
The roots of the indicial
equation are are p and p.
r(r 1) + r p2 = r 2 p2 = 0
Application of Frobenius Theorem yields a solution J p given by the formula
(1)n
J p (t) = t p t 2n
Consult the references for
the derivation of the formula for n=0
22n n!( p + 1)( p + 2) ( p + n)
J p (t).
The function J p (t) is called the Bessel function of order p of the first kind.
The series converges and is bounded for all t. If p is not an integer, it can
be shown that a second solution of Bessels equation is J p (t) and that the
general solution of Bessels equation is a linear combination of J p (t) and
J p (t).
For the special case p = 0, we get the function J0 (t) used in the aging
spring model in the second submodule of Module 11:
(1)n t 2n t2 t4 t6
J0 (t) = = 1 + +
n=0
(n!)2 2 4 64 2304
Note that even though t = 0 is a singular point of the Bessel equation of order
zero, the value of J0 (0) is finite [ J0 (0) = 1]. See Figure 11.2.
mathematicians believe), and where Hn = 1 + (1/2) + (1/3) + + (1/n) and is Eulers constant: =
0.5772 gives the first four digits. limn ( Hn ln n) 0.5772.
Id: chapter11.tex,v 1.10 1998-06-30 01:54:05-07 drichard Exp ODE Architect Workbook Page 210 on June 30, 1998 at 3:06
210 Chapter 11
1.25
0.75
0.5
0.25
-0.25
-0.5
-0.75
0 5 10 15 20 25
t
Figure 11.2: The graph of J0 (t) [dark] looks like the graph of the decaying
sinusoid 2/t cos(t /4) [light].
for arbitrary constants c1 and c2 . An important thing to note here is that the
value of Y p (t) at t = 0 does reflect the singularity at t = 0; in fact, Y p (t)
as t 0+ , so that a solution having the form given in (10) is bounded
only if c2 = 0.
Bessel functions appear frequently in applications involving cylindrical
geometry and have been extensively studied. In fact, except for the functions
you studied in calculus, Bessel functions are the most widely used functions
in science and engineering.
See Aging Springs in Bessels equation of order zero can be transformed into the aging spring equa-
Module 11. tion x + eat x = 0. To do this, we take
212 Chapter 11
Figure 11.3: Here are some typical graphs for the solution of x + C2 t2 x +
9x = 0 for various values of C2 . The graphs and the data tables are useful in
Problem 1 of Exploration 11.1.
Figure 11.4: Here is a phase-plane portrait for an aging spring ODE, x +
et x = 9.8. See Modeling an Aging Spring in the library folder Physical
Models and also Problem 1 in Exploration 11.3.
Id: chapter11.tex,v 1.10 1998-06-30 01:54:05-07 drichard Exp ODE Architect Workbook Page 213 on June 30, 1998 at 3:06
An optimal damping function b(t) is one for which the solution x(t) reaches
0.005 in minimal time t without ever going below zero.
C1 = 9 and t1 1.0897.
(a) Use ODE Architect to find an approximate optimal solution and values
of Ck and tk when k = 2. [Suggestion: Look at Figure 11.3.]
(b) Repeat with k = 3.
(c) Compare the optimal damping functions for k = 1, 2, 3, in the context
of the given physical process.
2. For
quadratic damping, b(t) = C2 t 2 , derive a power series solution x(t) =
n
n=0 a n t . Show that the recurrence formula for the coefficients is
[9an + C2 (n 1)a n1]
an+2 = , n1
(n + 1)(n + 2)
and a2 = 9a0 /2. Recall that a0 = x(0) and a1 = x (0).
Id: chapter11.tex,v 1.10 1998-06-30 01:54:05-07 drichard Exp ODE Architect Workbook Page 214 on June 30, 1998 at 3:06
4. If the robot arm is totally undamped, its position at time t is x(t) = cos 3t;
therefore the arm cannot reach x = 0 for all t, 0 t /6. In this situation
the undamped arm cant remain above x = 0. The optimal damping func-
tions Ck t k found in Problem 1 look more like step functions as the degree k
increases. Try to improve the time t by using a step function for damping.
Assume the robot arm is allowed to fall without damping until just before
it reaches x = 0, at which time a constant damping force is applied. This
situation can be modeled by defining
0 for 0 t < 6
b(t) =
B for t 6
for = 0.2, 0.1, and 0.05. Use ODE Architect to find values of B that give
an approximate optimal solution. Include a graph showing your best solution
for each and give your best value of t in each case. What happens to the
optimal B as 0?
5. Find a formula for the solution for the situation in Problem 4. The value of
should be treated as a parameter. Assume that x(t) = cos 3t for t < (/6) .
Then the IVP to be solved is
x + B x + 9x = 0
x(/6 ) = cos[3( 6 )] = sin 3
x (/6 ) = 3 sin[3( 6 )] = 3 cos 3
The solution will be of the form x(t) = c1 er1 t + c2 er2 t , r1 < r2 < 0, but the
optimal solution requires that c2 = 0. Why? For a fixed , find the value of
B so that x(t) remains positive and reaches 0.005 in minimum time.
Id: chapter11.tex,v 1.10 1998-06-30 01:54:05-07 drichard Exp ODE Architect Workbook Page 215 on June 30, 1998 at 3:06
2. Repeat Problem 1 for the functions Y0 (t) and t 2
sin t 4 . To graph a
good approximation of Y0 (t), solve the system equivalent of Bessels equation
of order zero (from Problem 1) with initial data t0 = 0.89357, x(t0 ) = 0,
x (t0 ) = 0.87942. As in Problem 1, we have to avoid the singularity at t0 = 0,
especially here because Y0 (0) = . The given initial data are taken from
published values of Bessel functions and their derivatives.
Id: chapter11.tex,v 1.10 1998-06-30 01:54:05-07 drichard Exp ODE Architect Workbook Page 217 on June 30, 1998 at 3:06
1. Check out the Library file Modeling an Aging Spring in the Physical Mod-
els folder (see Figure 11.4). The ODE in the file models the motion of a
vertically suspended damped and aging spring that is subject to gravity. Carry
out the suggested explorations.
2. Show that
t+1 3 3
x(t) = sin ln(t + 1) t + 1 cos ln(t + 1)
3 2 2
is an analytic solution of the initial value problem
x(t)
x (t) + = 0, x(0) = 1, x (0) = 0
(t + 1)2
Explain why this IVP provides another model for the motion of an aging
spring that is sliding back and forth (without damping) on a support table.
Id: chapter11.tex,v 1.10 1998-06-30 01:54:05-07 drichard Exp ODE Architect Workbook Page 218 on June 30, 1998 at 3:06
3. Graph the solution x(t) from Problem 2 over the interval 0 x 10 and
compare the graph to the one obtained in Module 11 using ODE Architect.
Id: chapter11.tex,v 1.10 1998-06-30 01:54:05-07 drichard Exp ODE Architect Workbook Page 219 on June 30, 1998 at 3:06
where is small enough that sin , the mass of the pendulum bob is 1,
and the value of the acceleration due to gravity is g = 32.1
1. With a = b = 1 and initial conditions (0) = 1 and (0) = 0, use ODE Ar-
chitect to solve ODE (13) numerically. What happens to (t) as t +?
2. Under the same conditions, what happens to the oscillation time of the pen-
dulum as t +? (The oscillation time is the time between successive
maxima of (t).)
1 See the article Poes Pendulum by Borrelli, Coleman, and Hobson in Mathematics Magazine,
Vol. 58 (1985) No. 2, pp. 7883. See also Child on a Swing in Module 10.
Id: chapter11.tex,v 1.10 1998-06-30 01:54:05-07 drichard Exp ODE Architect Workbook Page 220 on June 30, 1998 at 3:06
x-y
3
0
y
-1
-2
-3
-16 -14 -12 -10 -8 -6 -4 -2 0 2 4
x
Poincare map of a forced damped pendulum superimposed on a tra-
jectory.
Overview In this Chapter well look at solutions of a forced damped pendulum ODE. In
the linear approximation of small oscillations, this ODE becomes the standard
constant-coefficient ODE x + cx + kx = F(t), which can be solved explicitly in
all cases. Without the linear approximation, the pendulum ODE contains the term
k sin x instead of kx. Now the study becomes much more complicated. Well focus
on the special case of the nonlinear pendulum ODE
but our results leave a world of further things to be discovered. Well show that
appropriate initial conditions will send the pendulum on any desired sequence of
gyrations, and hint at how to control the chaos by finding such an initial condition.
Key words Forced damped pendulum; sensitivity to initial conditions; chaos; control; Poincare
sections; discrete dynamical systems; Lakes of Wada; control
See also Chapter 10 for background on the pendulum. Chapter 13 for more on discrete dy-
namical systems and other instances of chaos and sensitivity to initial conditions.
Id: chapter12.tex,v 1.3 1998-06-30 02:53:39-07 drichard Exp ODE Architect Workbook Page 222 on June 30, 1998 at 3:06
222 Chapter 12
Introduction
How might chaos and control possibly be related? These concepts appear at
first to be opposites, but in fact they are two faces of the same coin!
A good way to start discussing this apparent paradox is to think about
learning to ski. The beginning skier tries to be as stable as possible, with
feet firmly planted far enough apart to give confidence that she or he will not
topple over. If you try to ski in such a position, you cannot turn, and the only
way to stop, short of running into a tree, is to fall down. Learning to ski is
largely a matter of giving up on stability, bringing your feet together so as
to acquire controllability! You need to allow chaos in order to gain control.
Another example of the relation between chaos and control is the early
aircraft available at the beginning of World War I, carefully designed for great-
est stability. The result was that their course was highly predictable, an easy
target for antiaircraft fire. Very soon the airplane manufacturers started to
build in enough instability to allow maneuverability!
which arises when studying the stability of ships. The explorations at the end
suggest some strategies for these problems.
Lets begin to study ODE (1) with c = 0.1:
t-x
25
20
15
10
x
5
-5
-10
0 20 40 60 80 100
t
Figure 12.1: Solution curves of ODE (4) with x(0) = 0, x (0) = 2, 2.1.
Poincare Sections
Poincare found a way to understand and visualize the behavior of our differ-
ential equation: he sampled solutions of ODE (4) at multiples of the period
2 of the driving function:
0, 2, 4, . . . , 2k, . . .
This is much like taking pictures with a strobe light.
An equivalent way of saying this is to say that we will iterate1 the map-
ping P : R2 R 2 which takes a point (a, b) in R 2 , computes the solution
Note that the clock starts
at t = 0 when generating
x(t) with x(0) = a, x (0) = b, and sets
P(a, b) = (x(2), x (2))
0
Poincare plots. (5)
This mapping P is called a Poincare mapping. If you apply the operator P to
(a, b) k times in succession, the result is Pk (a, b) and we see that
Pk (a, b) = (x(2k), x (2k))
When the xx -plane is
used to chart the evolution of the
In sense, the Poincare section is simply a crutch: every statement about
Poincare sections corresponds to a statement about the original ODE, and vice
points Pk (a, b), k = 1, 2, . . . , it
is called the Poincare plane.
versa. But this crutch is invaluable the orbits of a nonautonomous ODE such
as (4) intersect each other and themselves in a hopelessly tangled way.
1 Chapter 13 discusses iterating maps f : R R; there you will find that already the map f (x) =
x(1 x) is filled with surprises. Before trying to understand the iteration of P, which is quite
complicated indeed, the reader should experiment with several easier examples, like linear maps R2
R2 . The notion of basin will also be much clarified by considering the iteration of Newtons method
in one complex variable, perhaps for cubic polynomials.
Id: chapter12.tex,v 1.3 1998-06-30 02:53:39-07 drichard Exp ODE Architect Workbook Page 224 on June 30, 1998 at 3:06
224 Chapter 12
Periodic Points
A good way to start investigating the Poincare mapping P (or for that matter,
the iteration of any map) is to ask: what periodic points does it have? Setting
x = y, a periodic point is a point (x, y) in R 2 such that for some integer k we
have Pk (x, y) = (x, y). Fixed points are periodic points with k = 1, and are
particularly important.
Periodic points of period k for P are associated with periodic solutions of
ODE (4) of period 2k. In particular, if x(t) is a solution which is periodic of
period 2, then
(x(0), x (0)) = (x(2), x (2))
is a fixed point of P. If you observe this solution with a strobe which flashes
every 2, you will always see the solution in the same place.
15
10
y
-5
-10
-15
-10 -5 0 5 10
theta (radians)
If you start the pendulum with both x(0) and x (0) small, the damping
will simply kill off the motion, and the pendulum will be attracted to the
downward equilibrium. The point (0, 0) in state space is called a sink.
The behavior is more interesting near an unstable equilibrium. Imagine
imparting an initial velocity to the bob by kicking it. For a small kick, it will
swing back. Now kick it a little harder: it will rise higher, and still swing
back. Kick it harder still, and it will make it over the top, and hit you in
the back if you arent careful. Dividing the kicks which dont make it over
from those that do is a very special kick, where pendulum rises forever, more
and more slowly, tending to the unstable equilibrium. Thus there are initial
conditions which generate solutions that tend to the unstable equilibrium; in
the Poincare plane these solutions form two curves which meet end to end
at the fixed point corresponding to the unstable equilibrium. Together they
form the stable separatrix of the fixed point. There are also curves of initial
conditions which come from the unstable equilibrium; together they form the
unstable separatrix of the unstable equilibrium. See Figure 12.3
As stated earlier, a good first thing to do when iterating a map is to search
for the periodic points; a good second thing is to find the periodic points which
correspond to unstable equilibria (saddles, in the case of the pendulum) and
find their separatrices.
For the unforced damped pendulum, the equilibria of the differential equa-
tion and the fixed points of any Poincare map coincide; so, too, do the separa-
trices of the unstable equilibria (in the phase plane) and the separatrices of the
corresponding saddle fixed points in the Poincare plane. These separatrices
separate the trajectories which approach a given sink from the trajectories that
approach a different sink.
Id: chapter12.tex,v 1.3 1998-06-30 02:53:39-07 drichard Exp ODE Architect Workbook Page 226 on June 30, 1998 at 3:06
226 Chapter 12
x-y
4
y
-2
-4
-10 -5 0 5 10 15 20
x
point of the boundary2 of any basin is in the boundary of all the others. Thus
if you start at a boundary point of any basin, and perturb the initial condition
an arbitrarily small amount, you can land in any of the infinitely many basins.
A careful look at Figure 12.4 should convince you that this stands a good
chance of being true: Each region of a canal boundary point includes pieces
on many curves. It isnt clear, of course, that there are canals of all the basins
between any two canals.
It is one thing to think that the Wada property is likely true, and quite
another to prove it. It isnt clear how you would prove anything whatsoever
about the basins: they do not appear to be amenable to precise study.
To get a grip on these basins, the first step is to understand why they
appear to be bounded by smooth curves, and to figure out what these smooth
curves are. For each sink (solid white squares in Figure 12.4), there are in fact
four periodic points (open squares), each of period two, which are saddles,
and such that for each saddle one of the two unstable separatrices is entirely
contained in the corresponding basin.
2 The boundary U of an open set U R2 is a point x R2 which is not in U, but such that
there exists a sequence of points xn U which converges to x. Later we will encounter the notion
of accessible boundary: the points x U such that there exists a parametrized curve : (0, 1] U
such that limt0 (t) = x. For simple open sets, the boundary and the accessible boundary coincide,
but not for our basins.
Id: chapter12.tex,v 1.3 1998-06-30 02:53:39-07 drichard Exp ODE Architect Workbook Page 228 on June 30, 1998 at 3:06
228 Chapter 12
The next step is to show that the accessible boundary of the basin is made
up of the stable separatrices of these saddles. This uses the technique of basin
cells, as pioneered by Kennedy, Nusse and Yorke. To see a fleshed out sketch,
see the C.ODE.E article referenced at the end of this chapter.
Gaining Control
The statement about the basins having the Wada property is, in some sense, a
negative statement, saying that there is maximum possible disorder. Is there
some positive statement one can make about the forced pendulum (for these
parameter values)? It turns out that there is. The precise statement is as
follows.
During one period of the forcing term, say during
t in the interval Ik = [2k, 2(k + 1)]
the pendulum will do one of the following four things:
It will cross the bottom position exactly once moving clockwise (count
this possibility as 1);
It will cross the bottom position exactly once moving counterclockwise
(count this possibility as +1);
It will not cross the bottom position at all (count this possibility as 0);
It will do something else (possibility NA).
Note that most solutions appear to be attracted to sinks, and that the stable
oscillation corresponding to a sink crosses the bottom position twice during
each Ik , and hence these oscillations (and most oscillations after they have
settled down) belong to the NA category.
The essential control statement we can make about the pendulum is the
following:
The chaos game in Module 12 suggests why this might be true; the tech-
niques involved in the proof were originally developed by Smale3 .
3 Stephen Smale is a contemporary mathematician who was awarded a Fields medal (the mathe-
matical equivalent of a Nobel prize) in the early 1960s. See Devaney in the references at the end of
this chapter.
Id: chapter12.tex,v 1.3 1998-06-30 02:53:39-07 drichard Exp ODE Architect Workbook Page 229 on June 30, 1998 at 3:06
Figure 12.5: Start in quadrilateral Q0 and reach forward into Q1 and backward
into Q1 .
230 Chapter 12
References Borrelli, R. and Coleman, C., Computers, Lies, and the Fishing Season in
The College Mathematics Journal, November 1994, pp. 403404
Devaney, R. L., An Introduction to Chaotic Dynamical Systems, (1986: Ben-
jamin/Cummings), Section 2.3 The Horseshoe Map
Guckenheimer, J. and Holmes, P., Nonlinear Oscillations, Dynamical Sys-
tems, and Bifurcations of Vector Fields (1983: Springer-Verlag). The
original classic in this field.
Hastings, S.P. and MacLeod, J.B., Chaotic Motion of a Pendulum with Oscil-
latory Forcing in The American Mathematical Monthly, June-July 1993
Hubbard, J.H., What It Means to Understand a Differential Equation in The
College Mathematics Journal, November 1994, pp. 372384
Hubbard, J.H., The Forced Damped Pendulum: Chaos, Complexity, and
Control in C.ODE.E Newsletter, Spring 95; (soon to be published in
The American Mathematical Monthly)
Sharp, J. A Problem in Ship Stability Group project for interdisciplinary
course in nonlinear dynamics and chaos, 1996-1997. Write to John Sharp,
Department of Physics, Rose-Hulman University, Terre Haute, IN.
Strogatz, S., Nonlinear Dynamics and Chaos, (1994: Addison-Wesley). Nice
treatment of many problems, including chaos induced in constant torque
motion.
Id: chapter12.tex,v 1.3 1998-06-30 02:53:39-07 drichard Exp ODE Architect Workbook Page 231 on June 30, 1998 at 3:06
Exploration 12.1.
1. Choose a value for c = 0.1, take A = 1 in ODE (1), and produce graphs like
those in the chapter cover figure and Figure 12.1.
2. Choose a value for A = 1 and c = 0.1 in ODE 1 and produce graphs like those
in the chapter cover figure and Figure 12.1.
5. Repeat Problems 1 and 2, but for the ODE with a quadratic nonlinearity,
x + cx + x x2 = A cos t
Id: chapter13.tex,v 1.7 1998-02-27 23:49:05-08 drichard Exp drichard ODE Architect Workbook Page 233 on June 30, 1998 at 3:06
400
300
x(n+1)=x(n)
x(n+1)
200
x(n+1)=f(x(n))
100
0
0 100 200 300 400
x(n)
Overview Processes such as population dynamics that evolve in discrete time steps are best
modeled using discrete dynamical systems. These take the form xn+1 = f (xn ),
where the variable xn is the state of the system at time n and xn+1 is the state of
the system at time n + 1. Discrete dynamical systems are widely used in ecology,
economics, physics and many other disciplines. In this section we present the
basic techniques and phenomena associated with discrete dynamical systems.
Key words Iteration; fixed point; periodic point; cobweb and stairstep diagrams; stability;
sinks; sources; bifurcation diagrams; logisitic maps; chaos; sensitive dependence
on initial conditions; Julia sets; Mandelbrot sets
See also Chapter 6 for more on sinks and sources in differential equations; Chapter 12 for
Poincare sections.
Id: chapter13.tex,v 1.7 1998-02-27 23:49:05-08 drichard Exp drichard ODE Architect Workbook Page 234 on June 30, 1998 at 3:06
234 Chapter 13
Equilibrium States
As with autonomous ODEs, it is useful to determine the equilibrium states for
a discrete dynamical system. First we need some definitions:
This use of the words A fixed point x of f is said to be stable (or a sink, or an attractor) if every
stable and unstable for points point p in some neighborhood of x approaches x under iteration by f , that
and orbits of a discrete system
differs from the way the words
is, if f n ( p) x as n +. The set of all points such that f n ( p) x
are used for equilibrium points of as n + is the basin of attraction of p. A fixed point x is unstable (or
an ODE. For example, a saddle a source or repeller) if every point in some neighborhood of x moves out of
point of an ODE is unstable, but a the neighborhood under iteration by f . If x is a period-n point of f , then
saddle point of a discrete system
is neither stable or unstable.
the orbit of x is said to be stable if x is stable as a fixed point of the map
f n . The orbit is unstable if x is unstable as a fixed point of f n . Stability is
determined by the derivative of the map f , as the following tests show:
Stable periodic orbits are attracting because nearby orbits approach them,
while unstable periodic orbits are repelling because nearby orbits move away
from them.
Id: chapter13.tex,v 1.7 1998-02-27 23:49:05-08 drichard Exp drichard ODE Architect Workbook Page 236 on June 30, 1998 at 3:06
236 Chapter 13
Figure 13.1: The cobweb diagram for the logistic map, xn+1 = 3.51xn (1 xn ),
suggests that iterates of x0 = 0.72 approach a stable orbit of period 4.
Is the fixed point in the chapter cover figure stable? Is the period-4 orbit
in Figure 13.1 stable? How about the two fixed points in that figure? [Sugges-
tion: Use the Discrete Tool as an aid in answering these questions.]
Figure 13.2: The cobweb diagram of the logistic map xn+1 = 3.9xn (1 xn ) sug-
gests that iterates of x0 = 0.8361 either approach an attracting periodic orbit
of very high period, or else wander chaotically.
238 Chapter 13
fixed point still exists and it is still attracting. The following theorem provides
a way of determining whether fixed points and periodic orbits are hyperbolic.
THEOREM 13.1 Given a discrete dynamical system x n+1 = f (xn ),
a fixed point x of f (x) is hyperbolic if | f (x )| = 1. Similarly, a peri-
odic point x of period n (and its orbit) is hyperbolic if |( f n ) (x )| = 1.
Because the number and type of periodic points do not change at pa-
rameter values where f (x) has hyperbolic points, we say that the qualitative
structure of the dynamical system remains unchanged. On the other hand,
this theorem also implies that changes in the qualitative structure of a family
of discrete dynamical systems can occur only when a fixed or periodic point
is not hyperbolic. We see this in the proportional growth model x n+1 = xn
when = 1 and = 1. For = 1 [and hence L 1 (1) = = 1 ] the
fixed point x = 0 is attracting. But for = 1 + the fixed point is repelling.
Thus, as passes through the value 1, the stability of the fixed point changes
from attracting to repelling and the qualitative structure of the dynamical sys-
tem changes.
Bifurcations
A change in the qualitative structure of a discrete dynamical system, such as
a change in the stability of a fixed point, is known as a bifurcation. Two other
types of bifurcations can also occur when f is nonlinear.
The first, known as a saddle-node bifurcation, occurs when x is a peri-
odic point of period n and ( f n ) (x ) = 1. In a saddle-node bifurcation, the
periodic point x splits into a pair of periodic points, both of period n, one of
which is attracting and the other repelling. A saddle-node bifurcation occurs
in the logistic growth family g (x) when = 1. At this value the fixed point
x = 0 (which is attracting for < 1) splits into a pair of fixed points, x = 0
(repelling for > 1), and x = ( 1)/ (attracting for > 1). This type of
bifurcation is sometimes called an exchange of stability bifurcation.
The second important type of bifurcation is called period-doubling and
occurs when x is a periodic point of period-n and ( f n ) (x ) = 1. In this
bifurcation the attracting period n point becomes repelling and an attracting
period-2n orbit is spawned. (Note that the stability can be reversed.) This
occurs in the logistic family g (x) when = 3. At this parameter value, the
attracting fixed point x = ( 1)/ becomes repelling and a stable period-2
orbit emerges with one point on each side of x = ( 1)/. Since the logistic
equations model population growth, this says that the population converges to
an equilibrium for growth rate constants less than 3. However, for values of
greater than 3, the population oscillates through a sequence of values.
The bifurcations that occur in a one parameter family of discrete dynam-
ical systems can be summarized in a bifurcation diagram. For each value of
the parameter (on the horizontal axis) the diagram shows the long-term be-
havior under iteration of a typical initial point. For example, if you see